Struct TRMConfig 

pub struct TRMConfig {
    pub hidden_size: usize,
    pub h_cycles: usize,
    pub l_cycles: usize,
    pub l_layers: usize,
    pub num_heads: usize,
    pub expansion: f32,
    pub pos_encodings: String,
    pub mlp_t: bool,
    pub halt_max_steps: usize,
    pub dropout: f32,
    pub vocab_size: usize,
    pub num_outputs: usize,
}

Configuration for Tiny Recursive Model

Based on TinyRecursiveReasoningModel_ACTV1Config from the Python implementation.

Fields

hidden_size: usize

Embedding/hidden dimension

h_cycles: usize

Number of high-level reasoning cycles

l_cycles: usize

Number of low-level update cycles per H-cycle

l_layers: usize

Number of layers in L-level (low-level) blocks

num_heads: usize

Number of attention heads

expansion: f32

FFN expansion factor (hidden_size * expansion)

pos_encodings: String

Positional encoding type: “rope”, “learned”, or “none”

mlp_t: bool

Use MLP instead of transformer (smaller, faster)

halt_max_steps: usize

Maximum steps for ACT halting

dropout: f32

Dropout probability

vocab_size: usize

Vocabulary size (for embeddings)

num_outputs: usize

Number of output classes/tokens

Implementations

impl TRMConfig

pub fn validate(&self) -> Result<()>

Validate configuration

pub fn head_dim(&self) -> usize

Get head dimension

pub fn ffn_hidden_size(&self) -> usize

Get FFN hidden size

Trait Implementations

impl Clone for TRMConfig

fn clone(&self) -> TRMConfig

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for TRMConfig

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for TRMConfig

fn default() -> Self

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for TRMConfig

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for TRMConfig

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.